Blog

In low and middle-income countries, many unique challenges remain to be overcome in pediatric medicine. As unique challenges require innovative solutions, Christopher Gill and his team at the Boston University School of Public Health (MA, USA) are pursuing many different avenues of investigation, with the overall aim of providing the greatest impact to the most vulnerable children and communities. 

For this blog post, Christopher describes a few key areas of research before discussing the importance of open research in the dissemination of information and ideas.

My name is Christopher Gill. I am a Clinician/Scientist and specialist in internal medicine and infectious diseases, with a career focus on child survival in low and middle-income countries. I have been a Professor in the Department of Global Health at Boston University School of Public Health since 2002, excluding a 3-year leave of absence during which I served as the Clinical Director for the Phase III Clinical Trials program at Novartis Vaccines and Diagnostics developing their CRM-197 conjugated Meningitis ACWY vaccine.  

A key research interest within my child survival work has been the epidemiology of respiratory infections and how programs and policies could be optimized to mitigate or prevent these diseases. This includes vaccine policy changes and strategies to optimize the diagnosis and case management of pneumonia. My research portfolio is broad, and I think will strike some as eclectic. But with a few notable exceptions, my projects are nearly all unified by the theme of child survival.

Investigating pediatric respiratory diseases to inform policy

Currently our team is working on five studies in parallel, which keeps us quite busy.

One of the most important projects is called the ‘Zambia Pertussis RSV Infant Mortality Estimation Study’ (ZPRIME), which is supported by the Bill and Melinda Gates Foundation. This is a large prospective post-mortem surveillance study that aims to measure the fatal impact of pertussis (whooping cough) and Respiratory Syncytial Virus (RSV), and the data from this project are meant to support policy decisions about whether the burden of fatal disease from either organism would justify new vaccines or other interventions at a global scale. We can report that RSV in particular appears to be a really important cause of mortality, detected in nearly 10% of deceased infants over the 3 years of the project, with most of that occurring infants <3 months of age who died in community settings outside of medical care. These are basic observations that don’t require much nuance to interpret but are extremely important in dictating how global health strategies and policies should be developed. 

In many ways, ZPRIME is one of the least complex projects we have ever done. The analysis itself is a basic n/N calculation where ‘n’ is deceased infants with a positive PCR, and ‘N’ is all deceased infants. But in other aspects, this is undoubtedly the most difficult project we have ever done. Emotionally it is devastating for our team who must consent and interview the bereft parents of these poor infants, day after day after day. Each time they provide an abbreviated grief counseling session to the family in an attempt to offer emotional support and empathy. After nearly 2,300 lost infants and counselling sessions, the cumulative weight and impact on our team who witness and experience this sadness, grief, and loss is incalculable. But the work is important and essential, and so we go on.

Evolving research projects to meet modern medical challenges

Since the start of COVID-19, the project evolved into a post mortem surveillance study for COVID-19’s impact in Zambia, and the initial results of this surveillance were published in January 2021 in the BMJ. Even though this was a side project nested within the main, this work has proved to be extremely impactful. By showing that COVID-19 deaths were common in a setting where they were not previously being reported systematically, we were able to shift the narrative from postulating exotic hypotheses to account for why COVID-19 seemed to skip Africa, to recognizing that this is almost certainly better explained by insufficient surveillance data. 

But there have been other bodies of investigation that emerged from this work as well. For example, in collecting information about the circumstances of each death that occurred in the community, we used a tool called a ‘verbal autopsy’. This is a systematic survey tool originally developed by the WHO to try and measure the impact of major syndromic causes of death in settings where resources or circumstances preclude an actual physical autopsy. Part of this tool is a free-text narrative section where the family member is asked to narrate in their own words the series of events that led to each child’s death. In reviewing the hundreds and hundreds of vignettes, we started to notice certain recurring themes emerging. One of these was the very high proportion of deaths where delays in seeking care appeared to have contributed to the fatal outcome. In effect, this was a source of qualitative data that, because it was collected systematically and comprehensively, could be quantified as well, something that is often impossible to do in qualitative research. 

In the process of sorting these deaths, we found that delays in care were the rule rather than the exception, occurring in over 80% of deaths, and occurring across the continuum of care: delays in deciding to seek care; delays in transit; delays while waiting at clinic to be seen; delays in referral to higher levels of care. It did not escape our notice the irony presented by these patterns: in the context of a study designed to inform the use of a very high tech and expensive set of interventions targeting a single pathogen (e.g., RSV), what we saw is how many deaths could be prevented in theory through low tech/low-cost interventions at the community level. We recently published these results in Pediatrics.

Similarly, we observed a high proportion of deaths that sounded very much like sudden infant death syndrome (SIDS) cases and/or accidental death due to smothering during sleep. SIDS has not been a major focus in global health research, and outside of a few studies in South Africa, there is not much known about this topic. And yet, there is no reason to assume that SIDS does not occur in Africa as it does everywhere else it has been investigated, and in the current context we were observing SIDS-like deaths in about 10% of all community deaths – not rare at all, and preventable. I currently have a doctoral student whose thesis is focused on exploring this further, to see what can be learned from analysis of the ZPRIME vignettes, while also collecting prospective data about the prevalence of known SIDS risk factors in this community, and qualitative data about how the community views the problem or its experience with SIDS-like deaths. In my opinion, this is an extremely important area to understand further, particularly as there are, again, very low tech/low-cost strategies for reducing SIDS deaths that could be acceptable and feasible in places like Zambia. 

Employing technology to approach unique pediatric challenges

Our group’s work has also included a number of technology-focused ongoing projects. The first is Project SEARCH (scanning ears for child health). SEARCH’s goal is to deliver a cell phone-based App designed to uniquely identify individuals through biometric analysis of ear morphology. And yes, ears are every bit as unique as fingerprints or retinal patterns or iris morphology, but far, far easier to access and image in young children.

The motivating interest behind this project is that in Zambia and places like Zambia, the provision of well-child care (including vaccines, HIV testing, growth monitoring, etc.) is coordinated through the use of paper well-child cards (often referred to as ‘under five cards’ or UFCs, since they are meant to last through the first five years of a child’s life). The advantage is that these cards are very accurate for uniquely identifying a child, but the downsides are devastating. The responsibility for maintaining these cards is solely on the parents/guardians of the child, and if they are lost or degraded to illegibility, there is no way to reconstruct these data. Moreover, their precious health data cannot be assessed, collated, and analyzed centrally through an electronic system, and so are useless for studying the implementation of well-childcare at a policy or planning level. An electronic medical record (EMR) is clearly a better solution. However, implementing an EMR is extremely challenging in situations where it is difficult to uniquely identify individuals over time and space through external identifiers, such as nation-wide ID cards. In other words, an EMR is not feasible unless the subject identification problem has first been solved. THAT is what we are trying to do, and I say that our results to data suggest that SEARCH could be a very powerful solution to that challenge.

One of our other ongoing technology-based projects focuses on portable bedside ultrasonography as a way of diagnosing pediatric pneumonia in settings where traditional X-ray is inaccessible. The devices in question are ultrasound transponders that connect to a tablet or smart phone and are about the size of an electric beard trimmer. The project has two objectives: 1) to assess the accuracy of portable ultrasound vs. X-ray; and 2) to use artificial intelligence (AI) to develop a machine learning algorithm to automate the interpretation of the ultrasound image sets, obviating the need for a radiologist to interpret them.

Objective 2 is the most challenging and interesting aspect to the project. The main challenge is that unlike X-ray, which presents a fairly stereotypical, static, single image, ultrasound data are collected as cinematic videos and involve multiple sweeps across the front sides and back of the left and right lungs – 12 sets of images per child, with thousands of individual image frames of data. To do this, we enrolled 200 children in Lusaka (Zambia) with suspected pneumonia, and obtained ultrasound and X-ray sets in parallel. From these we identified classic lobar pneumonia in about half, based on the X-rays. That yields about 100 children with consolidation patterns, but AI algorithms require thousands of images to train upon, making our sample size apparently woefully inadequate – except for the fact of fake-Vermeers.

What I mean by this is that the celebrated Dutch painter, Johannes Vemeer only created about 30 or so paintings, making him remarkably unprolific in comparison with say Rembrandt who created thousands of works. By contrast, the number of forged Vermeers are almost too many to count, but by studying the forgeries, one learns a lot about the actual masterpieces themselves. We used this same approach to expand our analytic sample size. Using something called a ‘GAN model’ (Generative Adversarial Network), we use the 100 ultrasound sets with clear consolidations to generate thousands of subtly altered ‘forgeries’ that display the same classic features as the original masterpieces, but allow us to expand our sample size to any size. We then use a combination of the ‘true’ and ‘forged’ image sets to train the AI model, and then, once trained, test the ability of the forgery-informed model to spot the true pneumonias. Currently, we are achieving >95% accuracy through this strategy. 

Suffice to say that we have managed to stay busy during this past year.

The importance of open research in the dissemination of information and ideas

The reason I became a scientist is that I love asking questions, answering them, pondering their implications, and then talking about this to whomever will listen. It is that last step that I think is so frustrating, because the prevalent publishing model creates immense barriers to the dissemination of information and ideas. In the US, science is largely funded by the US government, and as such is seen as being a national trust, an investment in all citizens. How then can we reconcile that by a journal refusing to allow access to the fruit of those investments – potentially publicly funded – as if somehow that information is now proprietary and privileged to those who can pay? That philosophy offends me to the bone, and that is why I think Open Access is so important.

You can read more about Project SEARCH with this Research Article, published with Gates Open Research – Project SEARCH (Scanning EARs for Child Health): validating an ear biometric tool for patient identification in Zambia >>